Inductive Logic Programming for Bioinformatics in Prova

This paper describes the inductive logic programming (ILP) features of Prova, a state-of-art distributed Semantic Web and Life Science inference service system and architecture for multi-relational data mining of complex Life Science phenomena such as complex biological relationships. The proposed novel design artifact implements typical ILP inference formalisms for rule-based generalization and specialization and combines them with expressive logic-based formalisms such as scoped meta-data based reasoning and typed logic in order to constrain the search space and the level of generality of relevant background knowledge. The tight integration of declarative rule-based programming with object-oriented programming (Java) allows outsourcing of computation intensive functionalities such as aggregations and data selections to highly optimized procedural code and query languages such as SQL, XQuery, OWL2Prova RDF, SPARQL. Parallel processing of ILP tasks is supported by a distributed service-oriented and event-driven middleware where several Prova rule engine instances are deployed on the Web as distributed inference services having access to modular data sources and distributed web-based resources. As a result our approach preserves the high expressiveness and flexibility of ILP for multi-relational data mining and attempts to overcome well-known computational and logical problems of ILP when facing very large and scattered heterogenous amounts of data with complex relationships published on the (Semantic) Web.